Using Machine Learning to Evaluate Biomechanics of Those Living with Multiple Sclerosis
Conference Year
January 2020
Abstract
Among the United States population, 2.3 million people suffer from Multiple Sclerosis (MS). MS is a neuro degenerative neurological disorder commonly characterized by fatigue, walking difficulties, numbness or tingling, and several other symptoms. These symptoms often pose risk to an individual's health, with a primary concern being fall risk. Out of every two Persons with Multiple Sclerosis (PwMS), one individual will fall in any given three-month period1, with half of these falls resulting in an injury that requires medical attention2. Given these statistics, it’s evident that preventative interventions are necessary for this population. Although risk assessments may be incorporated into bi-annual clinical visits, it’s evident these techniques do not instigate effective results regarding fall prevention. Clinical visits rather provide a means of generating a brief ‘snapshot’ of a patient's symptom progressive given sensory and motor impairment evaluations, lacking a more holistic view of daily symptom fluctuation.
With the progression of developing technologies such as wearable sensors, the ability to unobtrusively monitor metrics such as sway, gait, and balance is attainable in an accurate way of measure. Acknowledging these advancements, our goal is to develop a multi class activity classifier that is able to distinguish between several different daily living activities that can be applied to at home daily living data to generate information regarding biomechanic activity that can be used to predict fall risk using a fall risk algorithm generated using the in lab data set.
To accomplish this goal, data collected from 50 subjects instrumented with body-worn sensors while performing a variety of simulated daily activities and functional clinical tests within the laboratory environment and at home daily life for 48 hours will be utilized. In laboratory data will be used to develop the multi class activity classifier that can be implemented to at home data, providing a means of extracting activity data to be analyzed in an uncontrolled environment. Developing this set of data will enable biomechanical measures to be extracted from each occurence of each activity, capturing patterns in biomechanics and behavior that will be used to identify biomechanical and behavioral indicators of future falls during life. Our hypothesis is that this new generation of data will provide a larger data set that will improve and provide a more accurate assessment of a patient’s likelihood of falling in comparison to traditional clinical practices.
Primary Faculty Mentor Name
Ryan McGinnis
Faculty/Staff Collaborators
Ryan McGinnis (Collaborating Mentor)
Status
Undergraduate
Student College
College of Engineering and Mathematical Sciences
Program/Major
Biomedical Engineering
Primary Research Category
Engineering & Physical Sciences
Using Machine Learning to Evaluate Biomechanics of Those Living with Multiple Sclerosis
Among the United States population, 2.3 million people suffer from Multiple Sclerosis (MS). MS is a neuro degenerative neurological disorder commonly characterized by fatigue, walking difficulties, numbness or tingling, and several other symptoms. These symptoms often pose risk to an individual's health, with a primary concern being fall risk. Out of every two Persons with Multiple Sclerosis (PwMS), one individual will fall in any given three-month period1, with half of these falls resulting in an injury that requires medical attention2. Given these statistics, it’s evident that preventative interventions are necessary for this population. Although risk assessments may be incorporated into bi-annual clinical visits, it’s evident these techniques do not instigate effective results regarding fall prevention. Clinical visits rather provide a means of generating a brief ‘snapshot’ of a patient's symptom progressive given sensory and motor impairment evaluations, lacking a more holistic view of daily symptom fluctuation.
With the progression of developing technologies such as wearable sensors, the ability to unobtrusively monitor metrics such as sway, gait, and balance is attainable in an accurate way of measure. Acknowledging these advancements, our goal is to develop a multi class activity classifier that is able to distinguish between several different daily living activities that can be applied to at home daily living data to generate information regarding biomechanic activity that can be used to predict fall risk using a fall risk algorithm generated using the in lab data set.
To accomplish this goal, data collected from 50 subjects instrumented with body-worn sensors while performing a variety of simulated daily activities and functional clinical tests within the laboratory environment and at home daily life for 48 hours will be utilized. In laboratory data will be used to develop the multi class activity classifier that can be implemented to at home data, providing a means of extracting activity data to be analyzed in an uncontrolled environment. Developing this set of data will enable biomechanical measures to be extracted from each occurence of each activity, capturing patterns in biomechanics and behavior that will be used to identify biomechanical and behavioral indicators of future falls during life. Our hypothesis is that this new generation of data will provide a larger data set that will improve and provide a more accurate assessment of a patient’s likelihood of falling in comparison to traditional clinical practices.